How does Neural Architecture Search (NAS) apply to unsupervised learning?

NAS automatically discovers optimal architectures using unsupervised objectives instead of labeled validation loss.

In unsupervised NAS, architectures are evaluated using proxy objectives like reconstruction error, contrastive loss, or clustering quality metrics. Reinforcement learning, evolutionary algorithms, or gradient-based methods explore architecture space. This allows discovering encoders, decoders, and embedding networks optimized for representation learning tasks without labels.